Intelligent Star Wars Droid

Still working on the RL policy, I have Claude helping me bug fix the scripts for the training algorithms, its rough so far but I think if I optimize for low torque and then continuously lower kp I can make it work with my current constraints, so far I have moved the batteries in the head to increase stability and a better COM, but a working walking policy is still on the horizon

RL training in progress, you cant exactly see it in this screenshot, but I have hundreds of millions of robots being trained digitally with consideration of position, heat and balance of course. Now that I have a standing model, I will need to make some tweaks, like making sure it stands a bit taller so the shins dont collide with th etorso and implementing a walking policy.

New Spring system for balancing torque on knee, which will save energy, add more deep squat margin, and increase efficiency. Scoped spring and cord dimensions, learned how to auto implement screws in onshape, and got it all figured out for the most part, now all I need to do is lower weight, probably by decreasing batteries, and maybe I wont even have to reduce weight with this new spring in place, then I can implement the foam inserts on feet, remodel head, and maybe add head yaw, a lot to go but a majority of the work is done.

full body mockup, with brackets, major electrical components, actuators and timing belt/gear system installed. Now all that needs to be done is add masses and rotation limits and small tweaks, then I can simulate in mujoco, test for major flaws, onshape to urdf file, then to Isaac Lab.

Bracketing of electronics and added new antenna servos along with capstan drive for the antenna servos

Gear Synch Added! Ok so boom, gears/pulleys for the timing belt system are added and implemented, found all the parts I need on Amazon for ~$25 extra for both legs and used the provided seller info to spec the CAD models. Now, just have to add the foot, along with its connection to the ankle gear, and add all the significantly weighted tech pieces/ anything that takes up significant space to figure out bracketing. After that, add weights, clean rough edges, and train the RL policy while we wait for a grant.

New design work, made the motor proxies bigger to account for drivers, made thigh brackets bigger, added antennas with rotation axis, turns out im going to be using a dual antenna system, where only one is powered at a time based upon which garners a greater signal, but for right now only the antenna proxies are needed for weight calculations, fixed some more design details, am currently working on implementing belt system into cad.

Details Galore, added details to the thigh bracket, torso and head. Reproportionized a lot of stuff, all in all just a big design change day. Really hit a wave of frustration today when looking at some other people’s robots and how far other people have come. But then I started realizing that to compare to robots with different functions and different budgets is apples to oranges. I will continue to do the best with what I have. The budget has run out of control. Currently, my estimated cost to finish the robot is around an additional 350 dollars. I know that sounds rough but that’s where I’m at. Maybe GDZ34 drivers wouldnt be too bad a present for my birthday lol. All in all, rough day. I think I just need to fall in love with the design again and my motivation will come back. I plan on making the final-final version out of metal, so I will model for that.

NEW Mechanisms in the pipeline, removing ankle servo in favor of utilizing a chain linkage to the hip actuator to sync hip and ankle movements inversely. New designs, new CAD model for the foot. New Shin redesign. Due to heat strain on hip roll servos, a aluminum heatsink will be added, may require further cooling solutions in the future. Removing the ankle servos saves weight, distal mass and complexity of bus protocols. All in all, simpler is better and this new redesign is elegant and simple.

Final Touches, Moving to Sim————————Extended Shin, Recalculate Torque, Added Hip Roll Motor Horn Connectors, Signed up for AWS, requested more large server access (EC2), did material testing and found I only need ~avg 20% infill, made feet and imported to mujoco to get this awesome pic

CAD Leg Shenanigans– Found cheap ankle servos that work from Feetech, Remodeled the shin and thigh brackets for better aesthetics and new bracketing. To do: Find and implement hip roll servos, implement head tilt/pan servo (if able to- not super high priority) and make knee 3:2 ratio gear if needed.

Ive found it, the Goldilock of motors for my build. After hours of bartering with Alibaba factories, and messing with thigh and shin lengths, re-calculating torque requirements, I’ve found the motor that singlehandedly will solve all of my problems. The GIM3505-8 has RPM ~2x what I would have needed for walking, with enough torque for the hip pitch and possibly the knee pitch (hopefully). Because of the fact that my CAD and Mujoco models show most of the weight falling on the knee joint, as it is backward-facing, it is often holding the weight of the robot in solely its rotation. In addition, the knee joint requires the highest rpm and the highest range of motion. All these limits make it increasingly difficult to stay true to the form factor of the BD-1 without sacrificing money in the thousands. The motor comes with a encoder, so all I need is a driver. I thought about utilizing a all in one PCB board, with the Reinforcement Learning Policy, drivers and any other compute needed all in one place. Unfortunately, this elegant solution is very cost inefficent. While a great way to try my luck at PCB design, it will run me at least ~$30 more than just buying two arduinos and a raspberry pie 4. I still need to decide on the ankle motor replacement and the hip roll replacement as my old proxies were incorrect. (Claude doesn’t like CAD).

Shin + Thigh Re-proportioning and Motor selection and Motor CAD Implementation